Super-resolution for Bagadus

Related to 3D video there are a lot of challenges wrt. coding, processing and data transport. We are working experimentally on a streaming system that is installed in a sports stadium. You can see a video of our first demo at YouTube: http://www.youtube.com/watch?v=1zsgvjQkL1E

Goal

One application based on this streaming system would enable many spectators to see the replay of a particular event (e.g. a goal) in 3D, and zoom in to see whether a player was offside or not. Thus, it is desirable to play back the same scene from an arbitrary angle. This should also be possible inside the stadium during an ongoing match, where video streams must be combined to create scalable "free" viewpoint streams in real-time for recent events. Users may then replay the event multiple times from multiple angles (and even changing the angle during the playout). In such a scenario, there are multiple challenges like the 3D coding itself, data processing (video encoding) and delivery to 100s of concurrent users sharing the limited bandwidth inside the stadium.

How densely must we construct our camera array to use super-resolution for zooming in on individual players? Current sports broadcasting and analysis rely on camera operators who zoom in manually on interesting situations on the field. Trainer teams rely on camera teams who dedicate one camera to every player and follow them in zoomed-in mode throughout the entire game. It would be much more desirable if such zooming-in could be computed on demand from the recordings of static cameras, but camera resolutions don't provide the desirable level of detail. However, a series of papers has been published under the keyword super-resolution, where several picture are combined to create a new one of higher resolution. The number of input pictures for these algorithms is fairly large, and the goal of this thesis is to investigate if this technique can be applied in a feasible manner.

The thesis provides opportunities for active cooperation with the University of Tromsø and the University of Toulouse, including opportunities to visit both. The thesis requires a commitment to full-time studies, participation in meetings, regular presence in one of our labs, and interaction with other students.

Learning outcome

experiment setup, video coding, video compression, performance analysis

Qualifications

algebra, signal processing basics, programming

Publisert 28. juni 2016 10:31 - Sist endret 28. juni 2016 10:31

Omfang (studiepoeng)

60